Sundar Pichai Says Alphabet Can't Build AI Capacity Fast Enough, and Anthropic Has Secured 5 Gigawatts of It.

Alphabet said on its Q2 call that it remains supply constrained on AI computing capacity even after planning $195B–$205B of 2026 capex, highlighting strong demand. Google Cloud accelerated revenue growth to 82% YoY to $24.8B and more than tripled operating income to $8.8B (margin ~36%), while cloud backlog rose to ~$514B. The company is selling contracted multi-gigawatt TPU capacity to Anthropic for 2026–2027 and raised its capex guidance, with Q2 capex of $44.9B exceeding operating cash flow by $5.8B and funding supplemented by $49.6B of stock sales plus $20.3B of senior notes.

Analysis

The investment story is less about AI demand being strong and more about who owns the bottleneck. That favors GOOG/GOOGL because scarce compute can be monetized twice: first as a high-margin external service, then as a multi-year contracted annuity once capacity comes online. Second-order, this is mildly negative for the pure GPU trade if large customers increasingly mix TPUs/custom silicon into their stack, while AVGO benefits as the enabling semiconductor partner rather than just a commodity vendor.

Near term, the market may punish the capex intensity before it rewards the backlog. If free cash flow stays negative after heavy buildout, the stock can trade like a utility under construction, where multiples stay capped until investors see conversion from signed demand into durable cash generation. Over 1-3 months, the key catalyst is whether management shows that AI capacity is translating into operating leverage; over 6-18 months, the question is whether this becomes a structurally higher-ROIC platform or a perpetual reinvestment cycle.

Contrarian take: consensus may underweight the value of being supply-constrained, because scarcity is pricing power. But it may also be overconfident in backlog quality if a single fast-growing customer slows or if AWS/Azure/Oracle add enough competing capacity to erode the premium. The thesis is falsified if cloud margin rolls over materially or if capex steps up again without a corresponding improvement in cash conversion.

Key Decisions for Investors

• Long GOOG/GOOGL on 1-2% weakness for a 6-12 month hold; thesis is backlog monetization and cloud margin durability. Risk/reward: attractive if operating margin holds above ~30%; cut if capex rises again without FCF improvement.
• Buy a modest AVGO position as the higher-quality second-order beneficiary of TPU/custom-AI buildout over 6-18 months. This is a lower-beta way to own AI infrastructure, but trim if Google commentary suggests TPU demand is not broadening beyond one anchor customer.
• Do not short NVDA outright yet; instead, watch for evidence of TPU share gain or customer mix shift. If that appears, use a 3-6 month NVDA call spread hedge rather than a naked short, since secular AI demand can still overpower share loss.
• Relative-value idea: long GOOG vs. a cloud basket proxy (AMZN/MSFT) only if Google continues to show better AI monetization per dollar of capex. This works best on any post-earnings pullback and should be closed if peer cloud growth reaccelerates or Google backlog conversion disappoints.